LPR / Security Cam duty
- Thread starter pbc
- Start date
I don't think that is it since it was capturing these the night before. It's like ALPR just stopped from 8:34pm last night until 5:30 this morning. Trying to figure out why that would be since I don't think ALPR relies on schedules or whatever in BI, and I'm not even sure why it's putting squares around the motion since the LPR camera has "Do not highlight motion" set in BI (didn't even realize I had motion turned on!). Below is another it would have clearly captured but didn't.
biggen
Known around here
- May 6, 2018
- 2,692
- 3,066
Oh so the camera is working fine it appears but the problem is ALPR? Ok. Yeah I don't use any plate recognition software so I can't help with that unfortunately. I figure if I need to go find a plate I'll pull up the footage.
Robert G.
Getting comfortable
Those photos are a bit difficult to read. My eyes day maybe CCDN133?
My suggestion would be to conduct a test, I know you want to get three lanes of traffic but just for testing I would suggest zooming in. Concentrate on the single left lane, zoom in and capture only that lane if possible.
With it zoomed in just see how things work. You can always zoom back out and continue working at it, but at the distance I see in those photos even if the system can read those plates the number of incorrect reads would be high (my assumption).
Zoom in a conduct a couple days of testing, then go from there based on the results.
My suggestion would be to conduct a test, I know you want to get three lanes of traffic but just for testing I would suggest zooming in. Concentrate on the single left lane, zoom in and capture only that lane if possible.
With it zoomed in just see how things work. You can always zoom back out and continue working at it, but at the distance I see in those photos even if the system can read those plates the number of incorrect reads would be high (my assumption).
Zoom in a conduct a couple days of testing, then go from there based on the results.
Yeah, I have been, just seems odd that OpenALPR was capturing plates like crazy the day before and earlier in the day, then has just stopped. Even this morning it's very intermittent when it capture a plate, whereas when I started using it it literally captured almost everything in the day (I was actually trying to see if there was a way to schedule it in OPENALPR to shut off capturing plates during the day!).
Robert G.
Getting comfortable
There is no scheduling system that I have seen, I would actually like to shut off the system at night (I have no IR lights).
You can monitor the log file and see what it says. Watch when a vehicle drives past and see what it says. The logs have a constant display of information, like frames being processed, movement % etc. I think there are three values.
You also need to watch the CPU usage. If your CPU usage is high, it will have to drop frames - you can see that in the logs. I belive you have the GPU running, so CPU should not be an issue but something to watch for.
You can monitor the log file and see what it says. Watch when a vehicle drives past and see what it says. The logs have a constant display of information, like frames being processed, movement % etc. I think there are three values.
You also need to watch the CPU usage. If your CPU usage is high, it will have to drop frames - you can see that in the logs. I belive you have the GPU running, so CPU should not be an issue but something to watch for.
Hmmm...I am having some issues with BI using 60% so it jumps quite a bit.
What is your alprd service running at in terms of CPU? Mine is around 7 to 10%, doesn't really move from that as it uses my NVidia GPU for processing.
This is a recent log as a car passed: Can see the "Data accepted for processing", but nothing happened with it.
What is your alprd service running at in terms of CPU? Mine is around 7 to 10%, doesn't really move from that as it uses my NVidia GPU for processing.
This is a recent log as a car passed: Can see the "Data accepted for processing", but nothing happened with it.
Robert G.
Getting comfortable
How much CPU is used depending on your processor obviously, my CPU is an older i5. When I used the GPU, my CPU hovered around 20% and did not change when a vehicle passed by.
So you have about 28 frames per second normally, then when there is motion you can see in the 3rd column it says 100%. So that means all frames were processed, if that last number was not 100% then it means it can't keep up and some frames would not be processed. It looks fine based on that.
How many vehicles is it missing? All of them or just a small %?
So you have about 28 frames per second normally, then when there is motion you can see in the 3rd column it says 100%. So that means all frames were processed, if that last number was not 100% then it means it can't keep up and some frames would not be processed. It looks fine based on that.
How many vehicles is it missing? All of them or just a small %?
Robert G.
Getting comfortable
Long shot, but did you mess with the settings for limiting/controlling plate size? There is a feature where you can tell the system what size your plates are, anything outside that range is ignored.
Also, have you defined any masks, where it would ignore plates in the images?
Also, have you defined any masks, where it would ignore plates in the images?
No on the first, for the second I have masked the time stamp and camera name.
Tons of vehicles are missing, hasn't registered one in an hour and a half, and lots of folks going to work. Weird.
Maybe I'll delete the camera from OpenALPR and start over.
Tons of vehicles are missing, hasn't registered one in an hour and a half, and lots of folks going to work. Weird.
Maybe I'll delete the camera from OpenALPR and start over.
@pbc - Those plates are all clearly captured and identifiable, so I wouldn't mess with the camera focus and zoom now.
The problem is somewhere else.
Cross-threading - you are using the computer for a lot and doing AI and have been maxing out the CPU and I think you added a new graphics card last night.
I think the CPU is maxing out or ALPR isn't working right with that graphics card.
I would first do a reboot and see if that fixes it.
If not, start with the card you just added.
If not, maybe turn the AI off as that seems to be a big CPU draw.
Someone suggested you check the Limit Decoding box in BI to bring down CPU - did you do that - if so that could be the problem right there.
And then start working backwards.
The problem is somewhere else.
Cross-threading - you are using the computer for a lot and doing AI and have been maxing out the CPU and I think you added a new graphics card last night.
I think the CPU is maxing out or ALPR isn't working right with that graphics card.
I would first do a reboot and see if that fixes it.
If not, start with the card you just added.
If not, maybe turn the AI off as that seems to be a big CPU draw.
Someone suggested you check the Limit Decoding box in BI to bring down CPU - did you do that - if so that could be the problem right there.
And then start working backwards.
Last edited:
Interestingly, Deepstack and the AITools processes take up very, very little CPU. Alprd uses ~6% or so when it's not capturing a plate and maybe 10% when it is as presumably most of the processing is going to the GPU.
BI itself is the issue for me as it is taking up ~50 - 60%, even though I've got a fairly decent CPU (i7-6700), and have set things up with the Optimize CPU guide, and I don't think my cams are taking up too much (4500 kB/S and 800 MP/s on average).
I've posted pics of all my settings here if anyone has suggestions in that thread!
ipcamtalk.com
BI itself is the issue for me as it is taking up ~50 - 60%, even though I've got a fairly decent CPU (i7-6700), and have set things up with the Optimize CPU guide, and I don't think my cams are taking up too much (4500 kB/S and 800 MP/s on average).
I've posted pics of all my settings here if anyone has suggestions in that thread!
CPU running at 100%?
So uninstalled Deepstack AI thinking maybe that was causing an issue, but ever since I added 2 additional cameras to my Dell 3620 Precision tower (i7-6600 with 32GB of RAM), as soon as I launch BI the CPU jumps to 100%, then stays from 95% to 100%. Any ideas why that would happen? I'm only at...
@pbc -
Your bitrates are WAY too high and are running close to what many believe is capacity for that machine. Then toss on the other things you are adding to the machine and you get a 100% CPU usage.
On a bright sunny day, my cams are each at 250 or lower.
MY LPR is at 500 now, drops down to 10's at night as I run VBR.
Try the substreams on every camera but the LPR - and cut the bitrate back on the LPR - you don't need to be running 10,000. The substreams will bring you down a lot.
From the wiki:
0-500 MP/s ----- 3rd-7th gen i5 desktop CPU (4 cores)
500-800 MP/s --- 3rd-7th gen i7 desktop CPU (4 cores + hyperthreading) or 8th-9th gen i5 desktop CPU (6 cores)
800-1100 MP/s -- i7-8700 (6 cores + hyperthreading) or i7-9700 (8 cores) or the K edition of either
1100-1500 MP/s - i9-9900K
1500+ MP/s ----- Then it varies. Choose an HEDT platform and use quad channel memory. For the CPU, pick something near the top of this chart. Of course, you could also choose a different VMS that has lighter system requirements than Blue Iris.
Your bitrates are WAY too high and are running close to what many believe is capacity for that machine. Then toss on the other things you are adding to the machine and you get a 100% CPU usage.
On a bright sunny day, my cams are each at 250 or lower.
MY LPR is at 500 now, drops down to 10's at night as I run VBR.
Try the substreams on every camera but the LPR - and cut the bitrate back on the LPR - you don't need to be running 10,000. The substreams will bring you down a lot.
From the wiki:
0-500 MP/s ----- 3rd-7th gen i5 desktop CPU (4 cores)
500-800 MP/s --- 3rd-7th gen i7 desktop CPU (4 cores + hyperthreading) or 8th-9th gen i5 desktop CPU (6 cores)
800-1100 MP/s -- i7-8700 (6 cores + hyperthreading) or i7-9700 (8 cores) or the K edition of either
1100-1500 MP/s - i9-9900K
1500+ MP/s ----- Then it varies. Choose an HEDT platform and use quad channel memory. For the CPU, pick something near the top of this chart. Of course, you could also choose a different VMS that has lighter system requirements than Blue Iris.
Last edited:
biggen
Known around here
- May 6, 2018
- 2,692
- 3,066
Yeah my LPR at H.265 and CBR of 5888kbps generally runs at around ~750kB/s. This one you may have to run higher if using H.264. I know @bigredfish runs his at ~8000kbps on H.264.
My other cams are VBR and run around ~350ish kB/s or lower. Need to tweak your fps and bitrate. What resolution are you running in? I don't run above 1080.
My other cams are VBR and run around ~350ish kB/s or lower. Need to tweak your fps and bitrate. What resolution are you running in? I don't run above 1080.
bigredfish
Known around here
Strange no one else spotted that! Couple questions...
Do I adjust bitrate in the camera and leave BI as is? What bitrate do folks suggest for non LPR cameras?
Lastly, should I use CBR or VBR and H264 or 265?
I think I have the cams set at the "highest" quality for video in the Hik cameras if I recall.
Will check when I'm back home.
LOL... I know I know. But I'm stubbornly trying to get 3 lanes then complaining about it.
biggen
Known around here
- May 6, 2018
- 2,692
- 3,066
Resolution, FPS, bitrate, etc.. is all adjusted from within each camera.
Some use CBR while others use VBR. I use CBR with H265. Others don't. Just have to test what works for you.
@pbc -
Adjust the bitrates and stuff in the camera. Leave BI as is as it will auto adjust.
Everything else comes down to preference.
H265 let's you run a lower bitrate and saves storage space. Some feel they see better images in H264.
My suggestion is to start at 4,000 and then see if the image deteriorates and then go up or down accordingly until you see an improvement or degrading. Each cam will have a different level - cams that have a lot of contrast in pictures - a lot of grass and a lot of concrete typically need higher. I have some down as low as 2048 and one as high as 8192.
But definitely use the Substream feature in BI for all but the LPR - the CPU savings will be dramatic.
Some feel that CBR is better and others argue VBR is better - again - comes down to your situation. Personally I saw no difference between the two in image quality, so I switched to VBR as it will drop way down when not needed. Even for my LPR cam that the car is in for less than a second, the VBR keeps up and gets me a plate.
For LPR, I am a fan of zooming in as much as possible to make the plates as large as possible, but given your ideal straight on angle, you can probably get away with it zoomed out a bit. If you are able to get majority plates at your zoom and get 3 lanes and OpenALPR identifies them, that is all that matters. But most do need more zoom.
Adjust the bitrates and stuff in the camera. Leave BI as is as it will auto adjust.
Everything else comes down to preference.
H265 let's you run a lower bitrate and saves storage space. Some feel they see better images in H264.
My suggestion is to start at 4,000 and then see if the image deteriorates and then go up or down accordingly until you see an improvement or degrading. Each cam will have a different level - cams that have a lot of contrast in pictures - a lot of grass and a lot of concrete typically need higher. I have some down as low as 2048 and one as high as 8192.
But definitely use the Substream feature in BI for all but the LPR - the CPU savings will be dramatic.
Some feel that CBR is better and others argue VBR is better - again - comes down to your situation. Personally I saw no difference between the two in image quality, so I switched to VBR as it will drop way down when not needed. Even for my LPR cam that the car is in for less than a second, the VBR keeps up and gets me a plate.
For LPR, I am a fan of zooming in as much as possible to make the plates as large as possible, but given your ideal straight on angle, you can probably get away with it zoomed out a bit. If you are able to get majority plates at your zoom and get 3 lanes and OpenALPR identifies them, that is all that matters. But most do need more zoom.